BOSTON, Oct. 20, 2026 (GLOBE NEWSWIRE) — A growing market of tools tells brands how visible they are in AI search, often as a single score that blends ChatGPT, Claude, Grok and Perplexity together. A new benchmark suggests that blend hides the most important part of the result: the four engines largely disagree about whom to recommend.

The Q4 2026 AI Visibility Index, released today by Crackle PR, asked each engine’s API the same 25 questions about choosing a B2B tech PR agency, with web search on, five separate times. Average top-five shortlist overlap was about 43% when the same engine answered again, versus 17% across different engines. Overlap is measured as shared agency names divided by all distinct names across the two lists (Jaccard similarity), not the fraction of five slots repeated. On every one of the 23 questions where all four engines named agencies, average same-engine overlap exceeded average cross-engine overlap.

For a marketer, that means a single score can say a brand is winning AI search when it is strong on one engine and nearly absent on another. Which answer a brand gets can depend on which engine its monitoring tool happens to weight most.

ONE AGENCY, FOUR VERDICTS

Edelman, one of the world’s largest PR firms, shows how wide the split runs. It scored 22.0 out of 100 on Claude and 20.3 on Grok, but 3.3 on Perplexity and 0.7 on ChatGPT. Averaged together, it lands in the second tier. Drop either Claude or Grok from that average and Edelman falls out of the top five.

Depending on the engine checked, the same agency could be told it leads AI search or barely registers.

NOT JUST NOISE. DIFFERENT PICKS.

The disagreement was not simply run-to-run variation. In a follow-up comparison, each engine repeated its own picks more consistently than randomly assembled lists did. They repeated different agencies.

A follow-up check, run after the filed comparison test and labelled as such, compared the engines against shortlists assembled at random from the agencies named for each question, preserving list lengths. Rival engines overlapped less than the random lists did (17% versus 20%). That pattern is consistent with engine-specific sets of recommendations; it does not establish why the engines differ.

“Brands are being handed one number for their AI visibility, and our data says that number is an average of engines that don’t agree with each other,” said Parry Headrick, founder of Crackle PR, which publishes the Index. “One engine puts an agency on the shortlist while another doesn’t mention it at all. A blended score can tell you you’re winning when you’re invisible to the assistant your buyer actually uses.”

WHAT THIS MEANS FOR ANYONE MEASURING AI SEARCH

Based on the findings, the Index recommends that brands and their agencies:

  • Report visibility engine by engine. A single blended score can hide a strong showing on one engine and a near-absence on another.
  • Ask more than once. Even repeat answers from the same engine averaged only 43% shortlist overlap, so one answer is a weak sample.
  • Ask vendors how they weight the engines. The mix of engines behind a score can materially change the result, even when the questions and underlying answers stay the same.

BY THE NUMBERS

  • 17%: average cross-engine top-five shortlist overlap for the same question, measured as shared names divided by all distinct names across both lists.
  • 43%: average overlap when the same engine was simply asked again, using the same measure.
  • 23 of 23: eligible questions where average same-engine overlap exceeded average cross-engine overlap (follow-up check).
  • 25.9 points: the same-engine versus cross-engine overlap gap (95% interval 22.7–29.5 percentage points, from 1,000 question-level cluster bootstrap resamples). The direction held for top-three lists, full lists, a size-insensitive overlap measure, and with Crackle PR removed.
  • 17% vs. 20%: cross-engine overlap versus length-preserving random lists from the same agencies (follow-up check; 95% interval for the difference −4.7 to −1.4 percentage points).
  • 22.0 vs. 0.7: Edelman’s score on Claude versus ChatGPT, on a 0–100 scale.

WHAT HELD STEADY

Averaged across all 500 answers, the overall standings were stable: the same two agencies stayed in the top tier no matter which single engine was dropped. Stable doesn’t mean accurate. The Index measures how reproducible the answers are and how much they depend on the engine, not which recommendations are best.

“We don’t claim to know why the engines differ, or that buyers act on these lists,” Headrick added. “We filed the comparison test before computing it and published every answer. Nobody has to take our word for it.”

THE Q4 2026 STANDINGS

Scores are a 0–100 weighted composite across all four engines, not mention percentages. The selected standings below use corrected table r3, dated October 7, with bootstrap plausible rank ranges. Tiers are presentation groupings, not claims of statistically distinct performance; uncertainty ranges overlap.

Agency Score (0–100) Plausible rank range
Tier 1
Walker Sands 20.3 1–5
Highwire 20.3 1–3
Tier 2
PAN Communications 13.8 2–9
Bospar 13.6 2–6
Edelman 11.6 3–9
Inkhouse 9.5 4–12


SCOPE AND LIMITS

The results cover one industry (B2B tech PR), 25 designed questions rather than a random sample of real searches, and API access with web search on. They don’t describe the consumer apps, other industries, or how buyers behave. The overlap analysis excludes 34 answers naming no agencies. The random-list comparison and the drop-one-engine check came after the filed comparison test and are labelled as follow-up analyses.

METHODS AND DATA

Protocol v1.11 was frozen before collection. Pinned models: gpt-6-astra, claude-sonnet-5, grok-4.3 and sonar. Answers were collected October 2–3, 2026. Advance score notices were sent October 7 to agencies matched to the outreach contact list; not every scored agency received a notice. All 500 raw answers, the scoring code, the filed comparison addendum, and its OpenTimestamps proofs are public under CC BY 4.0. Q4 version DOI: 10.5281/zenodo.23164968. Full methodology: https://www.cracklepr.com/ai-visibility-index/methodology

DISCLOSURE

Crackle PR produced this research and is one of the agencies in the benchmark. It appears in the archive but is excluded from the ranking. The central result holds with Crackle removed. Public methods and archived answers allow scrutiny but don’t remove the conflict of interest.

ABOUT CRACKLE PR

Crackle PR is a B2B tech PR agency founded in 2020. It publishes the AI Visibility Index quarterly as an open, reproducible benchmark.

Media contact: Alexandra Yeager, alex@cracklepr.com


Media gallery

About The Author
Crackle PR